In this paper, we present a challenging task of 3D segmentation of individual plant leaves from occlusions in the complicated natural scene. Depth data of plant leaves is introduced to improve the robustness of plant leaf segmentation. The low cost RGB-D camera is utilized to capture depth and color image in fields. Mean shift clustering is applied to segment plant leaves in depth image. Plant leaves are extracted from the natural background by examining vegetation of the candidate segments produced by mean shift. Subsequently, individual leaves are segmented from occlusions by active contour models. Automatic initialization of the active contour models is implemented by calculating the center of divergence from the gradient vector field of depth image. The proposed segmentation scheme is tested through experiments under greenhouse conditions. The overall segmentation rate is 87.97% while segmentation rates for single and occluded leaves are 92.10% and 86.67%, respectively. Approximately half of the experimental results show segmentation rates of individual leaves higher than 90%. Nevertheless, the proposed method is able to segment individual leaves from heavy occlusions.

Video tracking based biological early warning system achieved a great progress with advanced computer vision and machine learning methods. Ability of video tracking of multiple biological organisms has been largely improved in recent years. Video based behavioral monitoring has become a common tool for acquiring quantified behavioral data for aquatic risk assessment. Investigation of behavioral responses under chemical and environmental stress has been boosted by rapidly developed machine learning and artificial intelligence. In this paper, we introduce the fundamental of video tracking and present the pioneer works in precise tracking of a group of individuals in 2D and 3D space. Technical and practical issues suffered in video tracking are explained. Subsequently, the toxic analysis based on fish behavioral data is summarized. Frequently used computational methods and machine learning are explained with their applications in aquatic toxicity detection and abnormal pattern analysis. Finally, advantages of recent developed deep learning approach in toxic prediction are presented.

Machine vision-based automatic detection of marine organisms is a fundamental task for the effective analysis of production and habitat changes in marine ranches. However, challenges of underwater imaging, such as blurring, image degradation, scale variation of marine organisms, and background complexity, have limited the performance of image recognition. To overcome these issues, underwater object detection is implemented by an improved YOLOV5 with an attention mechanism and multiple-scale detection strategies for detecting four types of common marine organisms in the natural scene. An image enhancement module is employed to improve the image quality and extend the observation range. Subsequently, a triplet attention mechanism is introduced to the YOLOV5 model to improve the feature extraction ability. Moreover, the structure of the prediction head of YOLOV5 is optimized to capture small-sized objects. Ablation studies are conducted to analyze and validate the effective performance of each module. Moreover, performance evaluation results demonstrate that our proposed marine organism detection model is superior to the state-of-the-art models in both accuracy and speed. Furthermore, the proposed model is deployed on an embedded device and its processing time is less than 1 s. These results show that the proposed model has the potential for real-time observation by mobile platforms or undersea equipment.

This paper proposes a new method of pest detection and positioning based on binocular stereo to get the location information of pest, which is used for guiding the robot to spray the pesticides automatically. The production of agricultural cultivation in greenhouse requires of big quantities of pesticides for pest control. Pesticides application is a major component of plant production costs in greenhouse, and the excess in their applications have a great negative impact on the environment. A pesticide application is ideal if the spraying coverage is presented as evenly distributed over the whole plant canopy and, if the product application is correctly adjusted for minimizing the losses towards the soil or the environment. In this approach, the difference of color features between pest and plant leaves is extracted by the image segmentation to identify pest. According to the results of image segmentation and binocular stereo vision technique, the 3D position of the pest has been obtained. In the process of position locating, centroid-matching technique is adopted to displace the common object-matching. The formula based on binocular stereo vision to measure distance is revised, additionally.

We introduce a multifractal analysis for detecting the small-size pest (e.g., whitefly) images from a sticky trap in situ. An automatic attraction system is utilized for collecting pests from greenhouse plants. We applied multifractal analysis to segment action of whitefly images based on the local singularity and global image characteristics. According to the theory of multifractal dimension, the candidate blobs of whiteflies are initially defined from the sticky-trap image. Two schemes, fixed thresholding and regional minima obtainment, were utilized for feature extraction of candidate whitefly image areas. The experiment was conducted with the field images in a greenhouse. Detection results were compared with other adaptive segmentation algorithms. Values of $F$ measuring precision and recall score were higher for the proposed multifractal analysis (96.5%) compared with conventional methods such as Watershed (92.2%) and Otsu (73.1%). The true positive rate of multifractal analysis was 94.3% and the false positive rate minimal level at 1.3%. Detection performance was further tested via human observation. The degree of scattering between manual and automatic counting was remarkably higher with multifractal analysis ($R2=0.992$) compared with Watershed ($R2=0.895$) and Otsu ($R2=0.353$), ensuring overall detection of the small-size pests is most feasible with multifractal analysis in field conditions.